Karpathy argues that while AGI is likely a decade away, its emergence will be gradual, blending into long-term economic trends rather than causing sudden disruption. Current LLMs, despite their capabilities, suffer from cognitive deficits like low output diversity and poor integration skills, especially in complex domains such as coding. Reinforcement learning, though flawed and susceptible to reward hacking, remains a better path than imitation learning. Human learning—rich in internal reflection and creativity—contrasts sharply with today’s AI, suggesting future gains may come from smaller, more efficient models trained on higher-quality data. Self-driving stalled due to high safety and capital costs, unlike digital AI applications. Education, meanwhile, should be reimagined around maximizing insight and understanding, using AI not just as a tool but as a collaborator in explanation and discovery.